Stasz, C., & Brewer, D. J. (1999). Academic Skills at Work: Two Perspectives (MDS-1193). Berkeley: National Center for Research in Vocational Education, University of California.

Study Approach

This report's fundamental premise is that skills are multidimensional. The preceding sections argue that, in the absence of a more rigorous conceptualization of different types of skills, and well-developed measures that capture them, we need to look to other possible proxies for skills. Just as variables such as test scores, curriculum indicators, and years of schooling serve as measures of academic skills, students' participation in extracurricular and part-time work activities may serve as measures of non-academic competence such as communication and teamwork skills. What is the implication of viewing skills in this way rather than as one dimensional? In a narrow sense, the answer to this question is that previous research--particularly quantitative research--on the effect of academic skills on labor market performance may have been misleading. We outline this argument in this chapter and, in doing so, the motivation for our empirical analyses becomes clear. Non-technical readers can skip the formal steps without loss of understanding in a broader sense, though one does not need formal statistical knowledge to realize that the policy implications could be profound. Focusing simply on academic skill development in education reform may cause a crucial component in preparing youth for the labor market to be ignored, and, in turn, may lead to a misplaced emphasis on academics. If non-academic skills are important and developed in arenas other than academic coursework, schools may wish to fundamentally redesign their priorities around a broader array of activities. The fact that American teenagers already spend a significant fraction of their time in extracurricular activities and in working part-time in the labor market (to a much greater degree than in other countries) may be prima facie evidence of the value of such activities.

Suppose individuals possess a "bundle" of knowledge, skills, and aptitudes that they bring to the labor market. These skills may be acquired through formal schooling, formal and informal training on the job, and a wide variety of other experiences. For simplicity, let's label these skills (S) as academic (A) and non-academic (N). Both these multifaceted skills may be rewarded in the labor market. In other words, an individual's wage (W) is a function of both types of skill, as well as many individual and/or job characteristics (X), and a random component ().[21]

(1) Y_ij = X_ij + S_j + _ij

If (1) is correctly specified, then standard Ordinary Least Squares (OLS) regression models will yield unbiased estimates of b and g, the effects of a change in individual/job characteristics, and in skills, on wages, other things equal, respectively. In most cases, however, the researcher is unable to include S in his or her model. In fact, non-academic skills are ignored, and the model is estimated with only variables for academic skills included. In other words, the true model is given by (2) but the researcher incorrectly estimates (3):

(2) LOG WAGE = X + ₁ A + ₂N +

(3) LOG WAGE = * X + *₁ A +

Estimating (3) may yield biased estimates of the effects of both individual/job characteristics and academic skills on wages.[22] If characteristics in X and academic skills (A) are uncorrelated with excluded non-academic skills, then OLS estimation will yield consistent and unbiased estimates of and ₁. (OLS gives estimates of the variance of and ₁, however, which are biased upwards, thus preventing valid inferences about the estimated coefficients of either individual and family background variables or academic skills.) More likely, there will be a correlation between background characteristics and non-academic skills and between academic and non-academic skills. This means that standard OLS models will yield biased estimates of the effects of both other characteristics and academic skills on outcomes like wages.

Whether the effect of omitting non-academic skills from this kind of statistical model leads to biased coefficient estimates and, hence, faulty inferences for policy is an empirical issue. The magnitude and direction of the bias depends (as can be seen from footnote 23) on (1) the extent to which there is a correlation between academic and non-academic skills, and (2) the true relationship between non-academic skills and wages (₂). If, as seems plausible from prior research, non-academic skills can also have positive effects on wages (₂ > 0), the existence of a positive correlation between academic and non-academic skills suggests traditional model estimates will have overstated the importance of academic skills in determining wages. In other words, on the one hand, part of the estimated return to academic skills is in part the return to non-academics. On the other hand, if academic and non-academic skills are negatively correlated, then the effect of academic skills on wages may have been understated. Bias in either direction could clearly lead to misleading policy inferences, so it is important to empirically investigate these relationships.

The key determinant of whether traditional analyses such as those discussed above yield incorrect statistical and policy inferences depends on the relationship between academic and non-academic skills, and between non-academic skills and labor market outcomes.[23] In other words,

• What is the relationship of non-academic to academic skills? Do students typically possess both types of skills, or is there a trade-off between academic and non-academic skills?
• How do non-academic skills affect labor market performance? In particular, how are the results of traditional econometric analyses of the determinants of wages affected by more completely representing an individual's skill bundle?

Since no well-developed measures of non-academic skills exist, tackling both these questions is difficult. A second best solution is to use participation in extracurricular and part-time work as proxies for non-academic skills. This is what is attempted in the remainder of the chapter.

[21]The outcome here is typically the natural logarithm of the hourly wage rate. Other outcomes, such as job status and the quality of work experiences, are clearly important; however, since wages are the outcome most widely studied in this context, and the most easily observable outcome, we frame the discussion in this section around wages as the outcome of interest.

[22]As is well known, the coefficients are given by

* = + ₂(NN)^-1 (NX)
* = ₁ + ₂(NN)^-1 (NA)

[23]Of course, analyses of these questions must also be attentive to how the answers to these questions might vary systematically across groups. For example, is the relationship between academic and non-academic skills different for college-bound and non-college-bound students? Or do non-academic skills have a different payoff in different segments of the labor market? Further research is needed on these issues.

[24]HSB cognitive tests were given in vocabulary, reading, and mathematics. In order to parallel our analyses of NELS:88, we report results using the mathematics raw test score number correct. Results using a valid composite score calculated by NCES that combines a student's score on each of these tests does not affect the results presented.

Stasz, C., & Brewer, D. J. (1999). Academic Skills at Work: Two Perspectives (MDS-1193). Berkeley: National Center for Research in Vocational Education, University of California.